The proposed method introduces an end-to-end training approach for a singular CNN, designed to concurrently handle star detection and centroiding. This method surpasses the centroiding accuracy and detection robustness achieved by several existing techniques.
The authors generate synthetic star images, augmented with real sensor noise and stray light, for training purposes to reduce the reliance on manual labeling efforts. A comprehensive performance evaluation of various CNN models is conducted, identifying the most suitable CNN architectures for real-time star tracker image processing.
The CNN-based method outperforms traditional star detection and centroiding algorithms in both synthetic and real-world tests, exhibiting superior resilience to high sensor noise and stray light interference. An additional benefit is that the algorithms can be executed in real-time on low-power edge AI processors.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Hongrui Zhao... at arxiv.org 05-01-2024
https://arxiv.org/pdf/2404.19108.pdfDeeper Inquiries